Weisskopf1983

Hearts and Minds: A Social Model of U.S. Productivity Growth
Author(s): Thomas E. Weisskopf, Samuel Bowles, David M. Gordon, Martin Neil Baily, Albert
Rees
Source: Brookings Papers on Economic Activity, Vol. 1983, No. 2 (1983), pp. 381-450
Published by: The Brookings Institution
Stable URL: http://www.jstor.org/stable/2534294
Accessed: 26/10/2010 17:10
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=brookings.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The Brookings Institution is collaborating with JSTOR to digitize, preserve and extend access to Brookings
Papers on Economic Activity.
http://www.jstor.org

THOMAS E. WEISSKOPF
University of Michigan
SAMUEL BOWLES
University of Massachusetts at Amherst
DAVID M. GORDON
New Schoolfor Social Research
Hearts and Minds:
A Social Model
of U.S. Productivfty Growth
Aftertenyearsof continuingimmersioninthewholeproductivityanalysisand
debate, whatcomes throughloudand clearis thatthereare some thingsthat
arecommonto all circumstancesof highlevelsofperformance.... Theseare
mattersof theheartandmindandnotof hardwareandcapital.-Hallett'
MOSTECONOMISTSagree that the slowdown in aggregate productivity
growthintheUnitedStatessincethemid-1960shasplayedapivotalrole
in the poor performanceof the U.S. economy. And yet, "despite
numerousstudiesoftheslowdown,"BPEAeditorsWilliamC. Brainard
Theauthors'nameshavebeenrandomlyordered.Theresearchwasoriginallylaunched
bya smallresearchgrantprovidedbytheProgressiveAlliance.WethankRezaBaharand
ClareBattistafortheirexcellentresearchassistanceandareespeciallygratefultoMichele
I. NaplesandJulietB. Schorfortheirhelpin modelingandconstructingsomevariables
uponwhichwe draw.We receivedvaluablecommentsfromparticipantsin the summer
1982conferenceoftheUnionforRadicalPoliticalEconomics,andfromtheWages,Prices,
andProductivityWorkingGroupattheEconomicsInstituteoftheCenterforDemocratic
Alternatives.AndrewGlynalsoprovidedveryusefulcommentsonanearlydraft.
1. JeffreyJ. Hallett,"Productivity-FromtheBottomUp," inRobertFriedmanand
William Schweke, eds., Expanding the OpportunitytoProduce:Revitalizing theAmerican
Economy through New Enterprise Development (Washington, D.C.: The Corporation for
EnterpriseDevelopment,1981),p. 406.
381

382 Brookings Paper-s oniEconomic Activity, 2:1983
and George L. Perryconclude, "its causes have remainedlargely a
mystery. '2
We think that this mystery stems more from the limitations of
conventionaleconomicanalysisthanfromtheimpenetrabilityof slower
productivitygrowth.Prevailingeconomicanalysistypicallyneglectsthe
humandimensionsof productionandthe institutionalcontexts within
whicheconomicactorsoperate.Wedevelopinthispaperanalternative
account of the productivityslowdown that addresses these various
lacunae.
We argue in particularthat declining work intensity and lagging
business innovation since the 1960s-factors that have been almost
entirelyelidedin recentanalyses-provide crucialmissingclues to the
productivitymystery.To developthisargumentwe presentandecono-
metricallytest a "social" modelof aggregateproductivitygrowth. It
integratestechnicalandsocialdimensionsofproductionandbuildsupon
ananalysisofthesocialsettingthathasconditionedproductivitygrowth
in the UnitedStates in the postwarperiod.3It can accountempirically
foralmostalltheproductivityslowdown.Ouranalysisis provisional;it
raisesmanyissuesforfurtherresearch,butwe believethatitprovidesa
promisingfoundationfor resolvingthe puzzle of slower productivity
growthinthe U.S. economy.
We presentan alternativeaccountof the productivityslowdownin
five parts.Some simple"stylizedfacts" relatedto workintensityand
business innovationare first summarized.We then outline the basic
elementsof a socialmodelof aggregateproductivitygrowth.A detailed
econometrictestofthatmodelisnextpresented,providingacomparison
ofitsexplanatorypowerwiththatofmoreconventionalapproaches.We
thenevaluateseveralpossibleadditionaloralternativehypothesesabout
the productivityslowdown, showingthatthe basicresultsof the paper
are robust even when confrontedwith competingor supplementary
interpretations.Thepaperconcludeswithabriefdiscussionofthepolicy
implicationsthatmightbedrawnfromourexplanationoftheproductivity
puzzle.
2. WilliamC. BrainardandGeorgeL. Perry,BPEA,1:1981,p. vii.
3. This paperdrawsheavilyon the historicaland structuralanalysisdevelopedin
SamuelBowles,DavidM.Gordon,andThomasE. Weisskopf,Beyondthe WasteLand:
A Democratic Alternative to Economic Decline (Anchor-Doubleday, 1983), especially
chaps.4-5.

T. Weisskopf, S. Bowles, and D. Gordon 383
DecliningWork Intensityand LaggingBusinessInnovation
Mosteconomistshaveconcentratedontwoprincipalexplanationsof
the slowdownin the rateof productivitygrowth-the oil-priceshock
andslowerratesof capitalformation.Butthese two explanationshave
two importantflaws:first,neitherfactortookeffect until1973-74,well
afterthebeginningof the productivityslowdownin the mid-1960s;and
second, even after 1973, these two factors appearto account for a
relativelysmallportionof theretardationinproductivitygrowth.
Journalists,businessobservers,andhistoriansfrequentlyarguethat
two otherseriousproblemsbeganto afflictthe U.S. economyafterthe
mid-1960s:friction began to mount at the workplace, leading to an
erosion of worker cooperationand worker effort; and corporations
turnedincreasinglytowardshorter-terminvestmentpolicies, resulting
inmoresluggishbusinessattemptsto improveproductiveefficiency.4
Thesetrends,ifmanifest,wouldobviouslyhelpexplaintheslowdown
inproductivitygrowthsincethemid-1960s.Buteconomistshavetended
to overlook these developmentseitherbecause they do not typically
studyworkintensityandbusinessinnovationinmacroeconomicanaly-
ses of aggregateproductivitygrowthorbecausethey suspectthatsuch
qualitativefactorscannoteasilybe integratedintorigorousquantitative
investigation.'
4. See, among many such recent commentaries,The Business Week Team, The
Reindustrialization of America (McGraw-Hill, 1982); James O'Toole, Making America
Work: Productivity and Responsibility (Continuum, 1981); and Ira C. Magaziner and
Robert B. Reich, Minding America's Business: The Decline and Rise of the American
Economy(Harcourt,Brace,Jovanovich,1982).
5. ThesequalitativefactorsaretheintangiblesaboutwhichC.JacksonGraysonwrites
in"EmphasizingCapitalInvestmentIsaMistake,"WallStreetJolurnal,October11,1982:
Concentrationon capitalinvestmenthas led to the relativeneglectof "otherfactors'
importantforgrowth-management,quality,technology,knowledge,employeeinvolve-
ment,trainingandlabor-managementcooperation.
Why?Forone thing,these otherfactorsaremostlyintangibles.Econometricmodels
neednumbers,andintangiblesaredifficulttomeasureandquantify.Also, . . . [it]is much
morecomfortableto workwiththingsyou can see, touchandkick. Forbothreasons,
theseintangiblesaremostoftenomittedfrommodels,policiesandmanagerialdecisions,
eventhoughcollectivelytheyhavealargerimpact.As theseotherfactorshaveincreased
in importance,theiromissionpartlyexplainswhy oureconomicpoliciesandforecasts
havebecomeincreasinglyinaccurate,andwhy ourproductivityslowdownhas been so
"puzzling"to many.

384 Brookings Papers on Economic Activity, 2:1983
It is impossible,indeed,to measuredirectlyor exactly eitherwork
intensityor innovativeactivity;one can neitherattachergometersto
workers in the shop and office nor track the frequency of creative
breakthroughsin corporateboardroomsandresearchlaboratories.But
we thinkthereis sufficientlycompellingindirectevidence of declining
work intensityand laggingbusiness innovationin the U.S. economy
since the mid-1960sto warrantcarefuland systematicintegrationof
these factorsintoanalysesof the productivityslowdownin the United
States.6
WORKPLACE FRICTION
From World War II through the early 1960s labor-management
relationsappearto havebecomeincreasinglypeacefulandcooperative.
By 1947,90percentofunioncontractsalreadypledgednostrikesduring
the term of contract. Strikeactivity itself declined substantially;the
proportionof worktimeidledbecauseof strikesfell, forexample,from
anaverageof 0.54percentinthefirstpostwarbusinesscycle, 1946-48,
to0.22percentinthenextfourcycles, 1948-66.Althoughtheearlydata
are somewhatfragmentary,it appearsthatthe proportionof workers
satisfiedwiththeirjobsalsoincreasedsignificantlyfromtheearly1950s
throughthemid-1960s.7
6. In all discussionthatfollows, we date businesscycles by choosingas peaksthe
yearsinwhichtheratioofactualtopotentialGNP,ascalculatedbytheBureauofEconomic
Analysisandrevisedby the Councilof EconomicAdvisers,reachedits business-cycle
peak.Whencomparingbusiness-cycleaverages,we datethecycles inthetextandtables
as extendingfrom one peak to the next even thoughmost of the cycle averagesare
calculatedfortheyearsextendingfromtheyearafterthepeaktothefollowingpeak.
7. Theunionno-strikecontractfigureisfromFredH.Joiner,"DevelopmentsinUnion
Agreements,"inColstonE. Warneandothers,eds., YearbookofAmericanLabor,vol.
2, Laborin PostwarAmerica(RemsenPress, 1949),p. 35. Strikefrequencyin this and
subsequentparagraphsis fromU.S. Departmentof Labor,Employmentand Training
Reportof the President,1981(GovernmentPrintingOffice, 1981),table G-8. Data on
trendsinworkdissatisfactionsummarizeresultsofaquestionaskedconsistentlysincethe
mid-1950sbytheOpinionResearchCorporation:"Howdoyoulikeyourjob-the kindof
workyoudo?"Wesummedthepercentagesresponding"verymuch"or"agooddeal."
Althoughthedataareconfidential,theyaresummarizedinMichaelR.Cooperandothers,
"Early WarningSignals:GrowingDiscontentamongManagers,"Business, January-
February1980.For dataonjob satisfactionreportedin subsequentparagraphs,dates
reportedinthetextcorrespondtotheparticularperiodsofaggregationreportedinthislast
article.

Realandobviousbenefitsto productionworkersduringthe postwar
boomled to thisharmony.Realspendablehourlyearningsincreasedat
anaverageannualrateof2.1percentfrom1948to 1966;andtheincidence
ofindustrialaccidents-one obviousbarometerofworkingconditions-
declinedby almostone-thirdfromits averageduringthe firstcycle of
the postwarboom, 1946-48, to its averageduringthe last cycle, 1959-
66.8
After 1965this rosy glow beganto fade. Duringthe next business
cycle from1966to 1973,forexample,theaverageannualgrowthof real
spendablehourlyearningsfellto 1.0percent,andtheaveragefrequency
of industrialaccidentsincreasedby 24 percentover its averagefor the
previouscycle.
At the same time, tight labor marketsincreasedworkers' relative
bargainingpowerandmoderatedtherisksofdismissal.Duringthe 1966-
73 cycle, for example,the averageratioof quitsto layoffsin manufac-
turingalmostdoubledits averagein the previouscycle, reflectingthe
greater sense of labor independencethat the late-1960s boom had
engendered.9
Takingadvantageof thisheightenedrelativeindependence,workers
appeartohavebecomeincreasinglyrestiveattheworkplaceandintheir
relationshipswith employers.The incidenceof part-timeabsenteeism
increasedby 32 percentfrom1959-66to 1966-73. The averageannual
percentageof worktime lost to strikesincreasedby 115percentfrom
1959-66 to 1966-71 before the Nixon wage-pricecontrols moderated
the surgein 1972-73.From1961-67to 1967-73,further,the percentof
strikesbecauseofworkingconditionsincreasedbymorethanone-third,
andthepercentof strikestakingplaceduringthetermof contract,more
familiarlyknownaswildcatstrikes,increasedby one-fourth.'0
8. Becauseof seriousproblemswiththetraditionalBureauof LaborStatisticsseries
on real spendableweekly earnings,we developedour own alternativeseries on "real
spendablehourlyearnings."See ThomasE. Weisskopf,"A New SpendableEarnings
Series,"TechnicalNote2(EconomicsInstituteoftheCenterforDemocraticAlternatives,
October1983).We similarlyresolvedproblemsin comparingaccidentratesfrombefore
andafternewreportingpracticeswereinstitutedin 1970.SeeMicheleI. NaplesandDavid
M.Gordon,"TheIndustrialAccidentRate:CreatingaConsistentTimeSeries,"Technical
Note 1(EconomicsInstituteoftheCenterforDemocraticAlternatives,December1981).
9. Quitsandlayoffs here andin subsequentparagraphsarefromEmploymentand
TrainingReport of the President, 1981, table C-14.
10. Dataon absenteeism,strikesto improveworkingconditions,andwildcatstrikes
arefromMicheleI. Naples,"TheStructureofIndustrialRelations,LaborMilitance,and

386 Br-ookingsPaper-s oniEconomic Activ'ity, 2.1983
This spreadingworker restiveness does not appearto have been
limitedtoblue-collarworkersorworkersinunions.Amongallproduction
personnel,bothblue-collarandwhite-collaremployees,job satisfaction
declined between 1965-69 and 1970-74. Accordingto detailed data
availablefromthe Universityof Michigan'sQualityof WorkLife Sur-
veys beginningin 1969,this decliningjob satisfactionwas surprisingly
widespread;itaffectedwhite-collar,professional,technical,andmanage-
rialworkersas wellas thoseinblue-collaroccupations."I
After1973,of course,labormarketsloosenedand-by conventional
expectations-workers' sense of independencewas boundto decline.
(The averageratio of quits to layoffs indeed declined, althoughonly
slightly, from 1966-73 to 1973-79.) But the restiveness apparently
persisted,despiterisingunemployment,andin manycases appearsto
havecontinuedspreading.Absenteeismratesdidnotdeclinein 1973-79
fromthe averageof the previouscycle, while both the percentageof
strikesover workingconditionsandthe percentageof wildcat strikes
increased.
Perhapsbecause workerswere still discontentedbut increasingly
fearfulabouteitherjob quits or proteststhroughstrikeactivity, they
appearto havebecomeincreasinglyalienatedonthejob after1973.The
theRateof Growthof Productivity:TheCaseof U.S. MiningandManufacturing,1953-
1977"(Ph.D. dissertation,Universityof Massachusettsat Amherst, 1982),tables 24
and4. Datesinthetextfor1961-66aredeterminedby 1961startingpointsonallthreedata
series.
Weconcentrateinthetextonaggregateindicatorsof trendsinworkintensitybecause
thefocusinthispaperisontheslowdowninaggregateproductivitygrowth.Werecognize,
nonetheless,thatsuchaggregativeindicatorsaresubjectto a widevarietyof distortions
andthatdisaggregatedindustrystudiesarenecessaryto providemoresubstantialsupport
forourhypothesesaboutlaggingworkintensity.Thetwo mostrigorousindustrystudies
of whichwe areaware,one on coal andone on automobiles,providestrongsupportnot
onlyforourinferencesaboutthe timingandmagnitudeof trendsin workintensityafter
the mid-1960sbutalso for ourhypothesesaboutthe linksbetweenthese developments
andthe industry-specificslowdownsin productivitygrowth.On the coal industrysee
M.Connerton,R. B. Freeman,andJ. L. Medoff,"ProductivityandIndustrialRelations:
The Case of U.S. BituminousCoal" (HarvardUniversity,Departmentof Economics,
December1979).On the automobileindustrysee J. R. Norsworthand C. A. Zabala,
"WorkerAttitudesandtheCostof Production,"paperpreparedfortheNationalBureau
of EconomicResearchWorkshopon InvestmentandProductivity,July 1983.Onmanu-
facturingasawholesee MicheleI. Naples,"TheStructureofIndustrialRelations,Labor
Militance,andtheRateof Growthof Productivity."
11. See thesequenceof surveys,Qualityof WorkLife(InstituteforSocialResearch,
Universityof Michigan),for1969,1973,and1977.

percentof nonsupervisoryworkerssatisfiedwiththeirjobs, according
to thelongestcontinuoussurveyavailable,fellfrom68percentin 1970-
74 to 59 percentin 1977-79,triplethe rateof declinefrom 1965-69 to
1970-74.Andthe pervasivenessofjob dissatisfactionbecameincreas-
inglyapparent.GrahamStainesconcludedfromthedetaileddataavail-
able in the Michigansurveysfrom 1973to 1977:"The sky hadfinally
fallen.Workersinvirtuallyalloccupationalanddemographiccategories
evidenced appreciable . . . [and] unmistakable manifestations of rising
discontent." 12
How canallthisevidencebe summarized?Wethinkthatmuchof it
canbe illuminatedthrougha simplepropositionthatwe develop more
formallyin subsequentsections: the higheris the cost to workersof
losing theirjobs, the more cooperative they are likely to be at the
workplace.Theloweris thecost oflosingtheirjobs,incontrast,theless
responsivethey will be to employereffortsto boost productivityand
extractgreaterlaboreffort.Wepresentinfigure1a summarymeasure
ofthe "cost ofjob loss," definedastheaverageannualpercentageof an
employee'slivingstandardthatarepresentativeworkercouldexpectto
lose if dismissed (see the text below for furtherdiscussion of this
variable).Superimposedontheannualseriesaretheperiodaveragesfor
1948-66, 1966-73, and 1973-79.Disregardingfor the momentcyclical
movements,the cost ofjob loss rose untilthe early 1960sandthenfell
precipitouslyuntilthe early 1970s;despitemuchhigherunemployment
ratesafter1973,the cost ofjob loss didnot returnto anythingclose to
itslevels duringthepostwarboom.
Based on this schematicandnecessarilyindirectevidence, we hy-
pothesizethatworkersbecameless fearfulof losingtheirjobs afterthe
mid-1960s,thattheybecameincreasinglyrestlessattheworkplace,and
thattheirlaboreffortmightwell have declinedas a result.Economists
mayhave been slow to recognizethese trends,butbusinessobservers
noticed them early. The Wall Street Journal reported in 1970, for
example:
Observersof the labor-managementscene . .. almostunanimouslyassertthat
thepresentsituationistheworstwithinmemory.. . . Moraleinmanyoperations
12. GrahamL. Staines,"IsWorkerDissatisfactionRising?"Challenge,vol.22(May-
June1979),p. 39.See alsoGrahamL. StainesandRobertP. Quinn,"AmericanWorkers
EvaluatetheQualityof TheirJobs,"MonthlvLaborReiiew, vol. 102(January1979),pp.
3-12.

388 Br-ookingsPapers oniEconomic Activity, 2:1983
Figure 1. The Cost of Job Loss, 1948-79a
Percent
40
30
20
1950 1955 1960 1965 1970 1975 1980
Sources: Text equation 6 and the data cited in the appendix.
a. The variable is defined as the average annual percentage of an employee's living standard that a representative
employee could expect to lose if that employee were dismissed from the job; the variable, expressed as a fraction,
has a potential range from zero to 1.0. The horizontal lines in the figure represent averages of annual data for 1948-
66, 1966-73, and 1973-79, respectively.
is saggingbadly,intentionalworkslowdownsarecroppingupmorefrequently
andabsenteeismis soaring.. . . [Manycorporations]contendtheproblem...
is so widespreadit's theirmajorheadacheatthemoment.'3
MANAGEMENT TORPOR
It has become almost commonplace among business observers, as
Business Week puts it, that U.S. corporations have recently suffered
"from a refusal to see beyond the next quarterly earnings statement." 114
Many suggest that corporations have been pursuing productivity-en-
hancing innovations less vigorously, shifting toward more speculative
financialinvestments and shrinkingfromthe longer-term entrepreneurial
risks that provide a dynamic impulse in a growing economy.
13. QuotedinJeremyBrecher,Strike!(StraightArrowBooks, 1972),pp.266-67.
14. The Business Week Team, TheReindustrialization ofAmerica, p. 48.

T. Weisskopf, S. Bowles, and D. Gordoni 389
Thesealarmsmaybe overblown,buttheremaynonethelessbe some
usefulkernelsamongthejournalisticchaff.Wethinkit quitelikely, on
the basis of availableand largelyindirectquantitativeevidence, that
U.S. corporationsgrew less andless inclinedto introduceproductive
innovationsafterthemid-1960s.
First,applicationsforpatentsgrewmoreslowlyas the boomturned
to stagflation.We can comparethe averageannualratesof growthin
patentapplicationsfiledfor inventionsin successive five-yearperiods
fromthemid-I950s:1956-60to 1961-65,2.5percent;1961-65to 1966-
70, 1.8percent;1966-70to 1971-75, 1.4percent;and1971-75to 1976-
80, - 0.1 percent.15
Second,by mostavailableinterpretations,thegrowthof privateand
publicexpenditureson researchanddevelopmentdeceleratedafterthe
late 1960s.Accordingto both KendrickandGriliches,for example, it
seems likely that the growthof R&Dexpendituresslowed duringthe
1966-73business cycle andthen slowedfurtheror perhapseven stag-
nated after 1973.16
Third,onecanobserveatrendafterthe 1960stowardgreaterrelative
use of corporatefundsforincreasesinfinancialassets-rather thanfor
realinvestmentand,therefore,forsupportorapplicationof productive
innovations.Increasesinfinancialassets,asapercentageofallcorporate
uses offunds,rosefromanannualaverageof 19.8percentin 1959-66to
25.4percentin 1966-73andto 25.8percentin 1973-79.'7
Asweargueinthefollowingsections,innovativepressureonbusiness
is best captured,otherthingsbeingequal,by changesin thefrequency
of business failures.Althoughthese failuresare obviously countercy-
clical, withthe deathsof firmsrisingwhenutilizationfallsinshort-term
contractions,we thinkthey arealso likelyto rise over the longerterm
15. Growthrates calculatedfrom data in U.S. Bureauof the Census, Statistical
Abstract of the United States, 1981 (GPO, 1981),table 945. The five-year periods reported
inthetextareconstrainedbythedatareportedinthesource.
16. See, forexample,JohnW. Kendrick,"Surveyof the FactorsContributingto the
Declinein U.S. ProductivityGrowth,"in FederalReserveBankof Boston, TheDecline
ofProductivityGrowth,ConferenceSeries22(FRBB,1980),pp. 1-21;andZviGriliches,
"R&DandtheProductivitySlowdown,"AmericanEconomicReview,vol. 70(May1980,
Papersand Proceedings,1979),pp. 343-48. See also KimB. ClarkandZvi Griliches,
"ProductivityGrowthand R&Dat the Business Level: Resultsfromthe PIMSData
Base," WorkingPaper916(NationalBureauof EconomicResearch,June1982).
17. Basedon dataon sourcesandusesof corporatefundsinEconomicReportof the
President,1981(GPO,1981),tableB-87.

390 Br-ookingsPapers on Economic Activity, 2:1983
whenbusinessinnovationis mostintense.Firmsthatdo notkeeppace
with or can least affordto keep up with modernizationwill be more
vulnerableto bothcollapseandbankruptcywhen the rateof business
innovationis high. Business failures do not cause intensificationof
innovativepressure,accordingto this argument,butthey are sympto-
maticofunderlyingincreasesintheforcesthatspurinnovativeactivity;
inotherwords,thedirectionof causalityis frominnovativepressureto
businessfailuresandnotthereverse.
This suppositionis reinforcedby the patternof businessfailuresin
the postwarperiod. Contraryto manyexpectations,business failures
havenotrisenmonotonicallywithdecliningutilizationratesduringthe
1970s-although they have obviously soaredsince 1979as a resultof
higherrealinterestrates.They were muchlowerin the 1970s,indeed,
thanduringtheyearsof sustainedprosperity.Wepresentadecyclicized
index of the frequencyof business failuresin figure2. We have both
taken the residualsof a regressionof the failurerate on an index of
capacity utilizationand taken a three-yearmoving average of that
residualtohighlighttheseculartrends.Thefigureseemsconsistentwith
muchof the business literature:forces creatingbusiness failuresrose
steadily throughthe mid-1960sand then declined steadily until the
dramaticshiftinmonetarypolicyinOctober1979launchedinterestrates
into orbit.18
Thismeasureof innovativepressureonbusinessis indirect,as is the
evidence on laggingwork intensity. It is nonetheless suggestive and
moreor less consistentwiththe qualitativeandcasualobservationsof
the businesscommunity.Wefindit plausibleto hypothesizethatU.S.
corporationshave been less likely to pursueproductiveinnovations
sincethemid-1960sthantheyhadbeenbefore.Suchaflagginginclination
towardproductiveinnovationis likely to have contributedto the pro-
ductivity slowdown. As thejournalistWilliamGreiderpuckishlyob-
servedinarecentarticle,"WhentheHarvardBusinessReviewdiscovers
that there is somethingwrong in the executive suite, something is
wrong."19
18. See theappendixtothispaperfordefinitionsandsources.Wediscussbelowsome
of the problemswithusingthe business-failurerateforthese purposes;see the section
below, "A CompositeSocialModelof AggregateProductivityGrowth,"andnote51.
19. WilliamGreider,"TakingCareof Business,"Rolling Stone, December9, 1982,
p. 11.

T. Weisskopf, S. Bowles, and D. Gor-don 391
Figure2. AdjustedMeasureof BusinessFailures, 1949-81a
Cyclicallyadjusteddeviationsaroundthe mean
(failuresper 10,000enterprises)
40
20
- 20-
-40 -
-60 -
-80 I I I I I I
1950 1955 1960 1965 1970 1975 1980
Sources: Authors' estimates based on data from sources cited in the appendix.
a. The original measure of the business failure rate is expressed as the frequency per 10,000 listed enterprises.
We then regressed this measure on an index of capacity with annual observations from 1949 to 1979 (the period over
which our subsequent econometric work was carried out). We next took a three-year (and end-of-period) moving
average of the residuals from the regression, both to control for the short-term business cycle and to smooth the
cyclical fluctuations in the underlying measure. The values for 1949, 1950, 1980, and 1981 were based on predicted
rather than actual observed residuals from the regression equation.
Determinantsof AggregateProductivityGrowth
Conventionalanalysesofproductivitygrowthhavereliedonwhatwe
callatechnicalmodelofproduction:givenexistingtechnicalknowledge,
laborandnonlaborinputsareroutinelyandpredictablytranslatedinto
output.The prevailingconclusionderivedfromthis model is that the
recentandpersistentslowdowninaggregateproductivitygrowthresults
from the fact, in MartinNeil Baily's words, "either that the rate of
technologicalchangeisnowmuchslowerthanitwas,orthattheeffective

flows of capitaland laborservices have grownmore slowly thanthe
measuredquantitiesof theseinputs,orperhapsboth."20
We do not doubtthat such "technical" factors affect the level of
aggregateproductivity.Butwe do questionwhethersucha mechanical
modelofinput-outputrelationsinproductioncouldpossiblycapturethe
more complex social determinantsof aggregateproductivityand its
growth. We have thereforesought a more complete social model of
productivitythattreatseconomicactorsas socialbeings,aspeoplewith
aspirationsandinhibitions,withneedsandresentments,witheconomi-
callyimportantandpotentiallymeasurablereactionstotheirinstitutional
settingandits history.Webeginwitha separateanalysisof thefactors
affectingworkintensityandbusinessinnovationandthencombinethat
analysiswithhypothesesderivedfromthemorefamiliartechnicalmodel.
WORK INTENSITY: THE MARX EFFECT
Marxiananalysisof the laborprocess has builtupona self-evident
proposition:theintensityofhumanlaborinproductioncanvarygreatly.21
20. MartinNeil Baily,"ProductivityandtheServicesof CapitalandLabor,"BPEA,
1:1981,pp.1-2. Wehavenotprovidedasurveyofthemainstreamliteratureonproductivity,
givenitsextensivecoverageinearlierissuesofBPEA,butwehavefoundespeciallyuseful
Edward F. Denison, Accounting for Slower Economic Growth: The United States in the
1970s(BrookingsInstitution,1979);J.R.Norsworth,MichaelJ.Harper,andKentKunze,
"TheSlowdowninProductivityGrowth:AnalysisofSomeContributingFactors"BPEA,
2:1979,pp. 387-421;Kendrick,"Surveyof theFactors,"pp. 1-21;andBaily,"Produc-
tivityandtheServicesof CapitalandLabor,"pp. 1-50.
21. Forsomeof themostimportantanalyticcontributionsintherecentliterature,see
StephenA. Marglin,"WhatDo Bosses Do?TheOriginsandFunctionsof Hierarchyin
CapitalistProduction,"ReviewofRadicalPoliticalEconomics,vol.6(Summer1974),pp.
60-112; HerbertGintis,"The Natureof LaborExchangeandthe Theoryof Capitalist
Production," Review of Radical Political Economics, vol. 8 (Summer 1976), pp. 36-54;
JamesDevine andMichaelReich, "The Microeconomicsof ConflictandHierarchyin
Capitalist Production," Review ofRadical Political Economics, vol. 12(Winter 1981), pp.
27-45; and Samuel Bowles, "The ProductionProcess in a CompetitiveEconomy:
Walrasian,Neo-Hobbesian, and MarxianModels" (Universityof Massachusettsat
Amherst,Departmentof Economics,May1983).
Someinterestingparallelsexist betweenthisneo-Marxianliteratureandsomerecent
neoclassicaldiscussionsof hierarchyandinvoluntaryunemployment.See, forexample,
GuillermoCalvo, "Quasi-WalrasianTheoriesof Unemployment,"AmericanEconomic
Review, vol. 69 (May 1979, Papers and Proceedings, 1978), pp. 102-07; and Edward P.
Lazear,"Agency, EarningsProfiles,Productivity,andHoursRestrictions,"American
EconomicReview,vol. 71(September1981),pp.606-20.

Werefertothispotentialvariationinworkintensityasthe"Marxeffect"
andrelyon someof Marx'sinsightsintothe structureanddynamicsof
productionincapitalisteconomiesinthefollowinganalysisofthiseffect.
Employerswilltry,otherthingsequal,tominimizethecost inwages,
supervision,andotherexpendituresof a unitof workdone. Theirtasks
arecomplicated,inlargepart,becauseemployeeswillbelikelyto value
positivelyatleastsomenonworkactivitiesduringworktime-which by
definitionwillnotcontributetotheoutputofthefirm-and willtherefore
seek to workless intensivelythantheiremployersprefer.Thispresup-
position does not implythatworkis absolutelyunbearableor that all
employersapplydraconianmeasurestoexploittheirworkers.Itpresup-
poses, muchmoresimply,thatemployeesandemployerspursueobjec-
tives affectingemployeeworkactivitythatarenotperfectlycongruent
andarehencepotentiallyinconflict.
We can thereforeanalyzethe employers'minimizationproblemby
focusing on workintensity,denotedas L* and definedas the ratioof
total effective laborinputsappliedin productionto the total hoursof
production-workerlaborpower hiredby the firm. Because effective
laborinputsarepotentiallyvariable,givenpurchasedlaborhours,work
intensityis clearlyvariableaswell, suggestingthecrucialimportanceof
examiningthedeterminantsof itslevel andvariation.22
A criticaldeterminantof workintensityis the effectiveness of em-
ployer controlover employees. This depends, in turn, on three main
factors:the expectedcost ofjob loss, theprobabilityof beingdetected
if an employeeis performingat a level of intensitybelow management
expectations.andtheprobabilityof jobloss if thatpoorperformanceis
22. The analysisin the followingdiscussiondrawsprimarilyuponSamuelBowles,
"CompetitiveWageDeterminationandInvoluntaryUnemployment:A ConflictModel"
(Universityof Massachusettsat Amherst,Departmentof Economics, 1981);DavidM.
Gordon,"Capital-LaborConflictandtheProductivitySlowdown,"AmericanEconomic
Review, vol. 71 (May 1981, Papers and Proceedings, 1980), pp. 30-35; andJuliet B. Schor
andSamuelBowles, "Conflictin the EmploymentRelationandthe Costof JobLoss,"
WorkingPaper6 (EconomicsInstituteof the Centerfor DemocraticAlternatives,July
1983).See alsoGerryOster,"LabourRelationsandDemandRelations:A CaseStudyof
the'UnemploymentEffect,' " CambridgeJournalofEconomics,vol.4(December1980),
pp. 337-48;GeoffHodgson,"TheoreticalandPoliticalImplicationsof VariableProduc-
tivity," CambridgeJournalof Economics,vol. 6 (September1982),pp. 213-26; and
MicheleI. Naples, "ProductionIs HumanActivity:A Social-RelationsApproachto the
ProductivitySlowdown," WorkingPaper4 (Economics Instituteof the Center for
DemocraticAlternatives,1982).

detected.Theproductof these threefactorsis the expectedcost to the
employeeofworkingatless intensitythanisexpectedbymanagement.23
Thefirstofthethreeisthemostobvious.Sinceemployersincapitalist
economiesmaynotphysicallycoercetheirworkersor deprivethemof
their liberty, one powerfulinstrumentof employer influence is dis-
missal-the threatof whichwilldependonthecost to workersof losing
theirjobs. Thehigherthecost ofjob loss, the morefearfulworkerswill
be offailureto conformto employers'intentions.
Theprobabilityofdetectiondependsontheintensityofsupervision-
thatis, surveillanceanddirectionof productionworkersonthejob. The
probabilityof dismissalvariesinverselywiththeextentofformalpower
of the employeebasedon unionsandothercollectiveemployeeassoci-
ationsandthedifficultytheemployerwillhaveinreplacingtheworker.24
Themultiplicativerelationamongthesethreefactorsimpliescomple-
mentarity,as seemsreasonable:intensesupervisionwilldolittlegoodif
thereis no cost to loss ofjob oriftheprobabilityof dismissalis low, for
example,andcostlyjob-lossconditionswillhavelittleeffectifemployers
areincapableof detectinganddismissingthosedeservingpunishment.
These hypotheses suggestthe followingexpressionfor the factors
affectingworkintensity:
(1) E = J S *D(U, R),
EJ, Es, ED, EJS, EJD, ESD>0, Du < 0, DR> 0,
where
E = indexof theeffectivenessof employercontroloverworkers
J = expectedincomeloss resultingfromjob loss
S = ratioof supervisory-workerhourstoproduction-workerhours
D = indexoftheprobabilityofdismissalifdetectedworkingbelow
managementexpectations
23. See Bowles, "TheProductionProcessina CompetitiveEconomy,"fora formal
argumentaboutthemultiplicativenatureofthisrelationship.
Ourformulationhereattemptsto model"objective"or "material"factorsaffecting
workintensity,andthereforeavoidstreatingworkerattitudesas purely"exogenous."
WethinkthatthisapproachispreferabletothatofanalystssuchasNorsworthandZabala,
in"WorkerAttitudesandtheCostofProduction,"whoimplicitlytreat"workerattitudes"
as a fullyexogenousfactoranddo notseekto understandwhatdeterminesvariationsin
thatdeterminantof costs andproductivity.
24. Thelogicunderlyingthesecondoftheseconditionsis similartothatofthesearch-
theoreticanalysisof theconditionsaffectingemployers'decisionto hirea newemployee
orto continuesearching.

U = level of collective worker membershipin unions or other
employeeassociations
R = indexof conditionsaffectingtheease withwhichmanagement
believesit canreplacea dismissedworker.
Workermotivationislikelytoconstitutethesecondmajordeterminant
of work intensity. If there are substantialimprovementsin workers'
earningsorworkingconditionsthatresultfromincreasesintheirintensity
of work,forexample,workersmaybe moreinclinedto cooperatewith
efforts to boost productivity.Conversely, if they must accept lower
wages to foster investmentor sufferspeedupand hazardousworking
conditionsto permitproductivityincreases,they maybe morelikelyto
intensifytheir resistanceto employers'efforts to extract more labor
activity.
It seemslikely,then,that
(2) M = M (W!, B, M*), MW!,MB, MM*> 0,
where
M = indexof workers'motivation
W! = motivation-enhancingfactorsinworkers'earnings
B = indexof qualityof workingconditions
M* = vector of any exogenous factors that positively influence
workers'job satisfaction.
We can combineequations1and2 to forma compositeexpression
forthe determinationof workintensity.Despitethe obviouslimitation
thatvariationsinthelevelofworkintensitycannotbemeasureddirectly,
itthenbecomespossibletospecifytheformofthefunctionalexpressions
impliedby 1and2and,asshownbelow,tomeasure-albeit imperfectly-
theirseveralcomponents.Thisfurtherstep makesit possible to draw
inferencesaboutunmeasurablevariationsinworkintensityfromvaria-
tions infactorsthatarelikelyto cause workintensityto vary;thuswe
canincorporateanalysesofthefactorsdeterminingtheintensityofwork
intoa moregeneralmodelof thedeterminantsof productivity.
INNOVATIVE PRESSURE ON BUSINESS:
THE SCHUMPETER EFFECT
Mainstreamanalysesof the aggregateproductionfunctionimplicitly
assumea standardneoclassicalmodelof competition:with conditions

of perfect competitionpresumed,the forces of competition remain
constant and continuousfor the individualfirm. As the production-
possibilityfrontiermoves outward,firmsare forced to remainon the
frontier because of the unrelentingforce of continuingcompetitive
pressure.Wecallthistheassumptionofautomatictechnicaladaptation.
Marx and Schumpetersuggested a quite differentconception of
competition.Marxoriginallyarguedthatcompetition,whileunrelenting,
was morelike warfarethana harmonyof mutualexchanges;farfrom
actingas passiveprice-takers,firmsengagecontinuouslyin attackand
counterattack,inforayandretreat.25Schumpetersubstantiallyextended
these insights.He suggestedthatinnovativebreakthroughscreatetran-
sitorymonopolypowerandgeneratequasirentsto be collectedby the
innovator,so that subsequentlycompetitorsare compelledto imitate
those innovations.In later work on business cycles and long waves,
Schumpeterarguedthatthereareperiodicwavesofcorporateinnovation
and entrepreneurialenergy. These waves of innovationalso unleash,
accordingto Schumpeter,"gales of creative destruction."The firms
leastableto ridethewaves of innovationfallbehindorfail.26
These insights suggest that variationsin innovative pressures on
business substantiallyaffectthe level andgrowthof aggregateproduc-
tivity. Automatictechnicaladaptationto exogenouslygeneratedtech-
nicalprogressimplies,otherthingsbeingequal,a relativelysteadyrate
of adaptationto exogenously generated expansion of productivity-
enhancingknowledge.If, however,therearevariationsinthe pressure
on firmsto adopt new methodsfor improvingproductiveefficiency,
exogenouslygeneratedtechnicalknowledge,even if growingat a con-
stant rate, will not lead to a steady rate of increase in aggregate
productivity.Whencompetitivepressuresarehigh,otherthingsbeing
equal,productivitygrowthwill also increase.Whencompetitivepres-
suresslacken,productivitygrowthis alsolikelyto moderate.27
25. Forusefuldiscussionsof Marx'sconceptionof competitionsee AnwarShaikh,
"PoliticalEconomyand Capitalism:Notes on Dobb's Theoryof Crisis," Cambridge
Journalof Economics,vol. 2 (June1978),pp. 233-51; andWilliSemmler,Competition
andMonopoly(ColumbiaUniversityPress,forthcoming).
26. See Joseph A. Schumpeter, The Theory of Economic Development (Harvard
UniversityPress,1934);BusinessCycles,vol. 1(McGraw-Hill,1939);and"TheAnalysis
of Economic Change," Review of Economics and Statistics, vol. 17(May 1935), pp. 2-10.
27. Weshouldnoteas anaddendumthatouranalysisof innovativepressures(andof
businessfailures)maybe distinguishedfromtwo strandsof prevailingdiscussionin the

This hypothesis can be translatedinto a form that permits direct
incorporationinto a model of aggregateproductivitygrowth. Many
neoclassicalgrowthmodelssuggestthattechnicaladvancecontributes
to aggregateproductivitygrowthexponentiallyover time, reflectedin
the exponentialadjustmentfactor, eX'.Relying on what we call the
Schumpetereffect, we substitutea behavioralrelationfor the constant
parameterX.
We begin with the propositionthat the actualrate of implemented
technicalprogressisafunctionofboththeexogenousgrowthofpotential
technical progress and the currentlevels of innovative pressure on
business:
(3) A = A* + VLC, L> 0,
where
A = actualrateof technicalprogress
A* = rateofgrowthofexogenouslygeneratedtechnicalknowledge
11 = coefficientof afirm'sadjustmenttovariationsincompetitive
pressure
C = indexof the (variable)level of competitivepressureonfirms
to improveproductiveefficiency,witha meanof C equalto
zero.
This formulationsuggests that the neoclassical model is simply a
specialcase: ifcompetitivepressurewereconstant(andequalto C), the
businessandeconomicsliterature.Oneemphasizeschangesinbusinessattitudesorrisk
preferencesasasourceofvariationsininnovativepressure.Thisisespeciallycharacteristic
of recentjournalisticdiscussionsof managementfailuresinthe UnitedStates.A second,
followingthe Schumpeterianlead, placesemphasison purelytechnicaldeterminantsof
wavesof innovativepressure,focusingonburstsofinventionandsubsequentdiffusionof
"epoch-making"ideas. For a justifiablycriticalreview of this approach,see Edwin
Mansfield,"LongWavesandTechnologicalInnovation,"AmericanEconomicReview,
vol. 73(May1983,PapersandProceedings,1982),pp. 141-45.Weargueincontrastthat
longswingsininnovativepressurereflectandareconditionedbytheconstructionof new
social structuresof accumulationandthe long periodsof expansionthey support.We
concentrate,in otherwords, on structuralforces affectinginnovativepressuresrather
thanonchangesinattitudesortrendsintechnicalinventiveness.Forfurtherelaboration,
see DavidM. Gordon,RichardEdwards,andMichaelReich,SegmentedWork,Divided
Workers: The Historical Transformation of Labor in the United States (Cambridge
UniversityPress, 1982),chap. 2; and David M. Gordon,ThomasE. Weisskopf,and
SamuelBowles, "LongSwingsandthe NonreproductiveCycle," AmericanEconomic
Review, vol. 73 (May 1983, Papers and Proceedings, 1982), pp. 152-57.

actualrateof growthof technicalknowledgewouldbe equalto X*.But
if competitivepressurewere not constant, then C would vary, and X
wouldthusdivergefromX*.
These propositionsfurther suggest that variations in innovative
pressurehavea cumulativeeffect, augmentingorerodinga firm'sincli-
nationstoinnovateoveraperiodofyearsdefinedprimarilybythefirm's
planninghorizons and the average amortizationperiod of technical
innovations.Short-termcyclical variationsin innovativepressureare
less likelyto affectafirm'sbehaviorthansustainedtrendstowardmore
orless intensecompetitionoversubstantiallylongerperiods.Assuming
that we can find a reasonableproxy measurefor C and that we can
properlycumulateits effects over time, the Schumpetereffect should
bejust as susceptibleto analyticinvestigationas theMarxeffect.
TECHNICAL FACTORS AFFECTING PRODUCTIVITY
Our attention to the Marx and Schumpetereffects is in no way
intendedtodiminishtheimportanceofrelationshighlightedbythemore
familiartechnicalmodel of productivity.We expect that the level of
capacityutilization,the capitalintensityof production,andvariations
intherelativepricesof externalinputsallaffectaggregateproductivity.
Wenoteherethemaindirectionsof effect.
Capacity Utilization. There are several importantreasons for ex-
pectingcovariationbetweenaggregatelaborproductivityandcapacity
utilization.First,theremightbe hoardingof nonproductionworkersin
business-cycledownturns,whichwouldleadto reducedlaborproduc-
tivity (perpurchasedtotal laborhours)duringrecession. Second, and
similarly,therewouldbe underutilizationof ownedfixed-capitalstock
duringa downturn,whichwouldleadto decreasedactualuse of capital
inputsin production.Third,the efficiency of the productionprocess
itself-including theefficiencyofutilizedcapitalinputsandofthelabor-
managementapparatus-might be reducedif capacity utilizationfell
belowsometargetedorwarrantedlevelforwhichproductiveoperations
hadbeen designed.Fourth,it is conceivablethattheremightbe some
hoardingofproductionworkersduringthecycle, withexplicitorimplicit
contractspreventingtheimmediatelayoffsthatlowercapacityutilization
mightotherwise cause. A fifthfactor mighthave the opposite effect:
duringperiodsof low capacityutilization,firmsmightretiretheirleast

efficientcapitalequipmentandthusincreasetheaverageproductiveness
of capitalgoods stillinuse.
Inthesubsequentanalysiswecontrolforthefirsttwofactorsdirectly.
Hoardingof nonproductionworkersduringthe businesscycle is taken
intoaccountbyexpressingproductivityasoutputperproduction-worker
hour. Variableutilizationof the fixed capital stock over the cycle is
testedbyadjustingmeasuresoftheownedfixed-capitalstockforvariable
levels ofutilization.
The net effect of the last threefactors is impossibleto anticipatea
priori.Weweightherelativeimportanceofoffsettingpossibleeffectsof
lowercapacityutilizationby testingdirectlyforthe directionof covar-
iation,otherthingsequal,betweenaggregateproductivityandthelevel
of capacityutilization.Wecontrolforthepossibilityof sluggishadjust-
mentof staffinglevelstothebusinesscyclebypostulatingthataggregate
productivitywill also varydirectlywiththe rateof changeof capacity
utilization;themorerapidlyutilizationlevelsareincreasing,forexample,
the greateris the likelihoodthatfirmswill makeincreasinglyefficient
use of theresourcesandtheemployeesthatmighthavebeenpartlyidle
duringperiodsof contraction.These controlsareobviouslyroughand
imperfect,but they shouldat least allow testing of hypotheses about
otherpossibledeterminantsof aggregateproductivityby provisionally
controllingforthepossibleinfluenceofvariationsincapacityutilization.
Capital Intensity. Aggregate labor productivity clearly varies with
thecapitalintensityofproduction.Butaggregatecapitalintensitycannot
be measuredby the aggregateratioof the (valueof the) owned capital
stock to labor hours for at least two importantreasons. First, some
portionof theownedcapitalstockmaynotenterintoproduction;itlies
unutilizedas a result of fluctuationsin aggregateeffective demand.
Second,theremaybe significantvariationsin the averageefficiencyof
effective capital services, resultingprimarilyfrom the possibility of
obsolescenceinthecapitalstocknotcapturedby theusualadjustments
fordepreciation.
These considerationsclearlysuggestthe need for two independent
kindsofadjustmentstotheaggregatecapital-laborratio-one forvariable
levels of utilizationand anotherfor factors that may lead to variable
ratesofobsolescence.Wepursuebothkindsof adjustmentsinsucceed-
ingsectionsof the paper,althoughtheformeris mucheasierto specify
thanthelatter.

External Input Prices. Several economists, particularlyMichael Bruno
and Jeffrey Sachs, have recently focused on the effects of input price
shocks on productivity growth in the United States and other advanced
economies. Bruno states their conclusion quite simply:
For a raw-materialintensiveactivitythe conventionaltwo-factorview of the
productionprocessis onlyvalidwhenthe relativepriceof the rawmaterial(in
outputunits)orits unitinputstaysconstant.Whenits relativepricerisesandit
is a complementaryfactor of production,productivityper unit of the other
factors,labourandcapital,mustfall.28
We agree with the microeconomic presuppositions of this argument;
increases in external input-prices of crude and intermediate materials
are likely to reduce the productivity of labor and capital inputs. We are
concerned, however, about the way in which this insight is applied and
about the specification of relative input prices in the recent literature.
We make an effort in subsequent sections to improve upon the empirical
specification of this effect.29
A CompositeSocialModel of AggregateProductivity
It is now possible to combine both social and technical dimensions
into a composite model of aggregate labor productivity. We follow
convention and express productivity as a multiplicative function of the
several inputs and social factors that are likely to affect it.30 The basic
model becomes
28. MichaelBruno,"WorldShocks,MacroeconomicResponse,andtheProductivity
Puzzle,"WorkingPaper942R(NationalBureauof EconomicResearch,July1982),p. 2.
See also MichaelBruno, "Raw Materials,Profits,and the ProductivitySlowdown,"
WorkingPaper660(NationalBureauof EconomicResearch,April1981);andBrunoand
JeffreySachs,"InputPriceShocksandtheSlowdowninEconomicGrowth:TheCaseof
U.K. Manufacturing,"WorkingPaper851 (NationalBureauof EconomicResearch,
February1982).
29. We note for purposesof claritythat Brunoactuallyidentifiestwo alternative
possibleeffects of relativeexternalpriceincreaseson productivitygrowth.One is the
substitutioneffect, illustratedby the quotationin the text. The otherfocuses on the
measurementbiasresultingfromthestandardprocedureofdoubledeflationofgrossoutput
datainthenationalaccounts.As Brunonotes,itis difficultto separatethetwo.
30. Thecrucialassumptioninthisformulationthatcannotbeavoidedisthatofconstant
returnsto scale,andwecommentontherelevanceofandevidenceforthisassumptionas
theeconometrictestsproceed.See note54below.Foradetailedcritiqueofthetraditional
inferencesaboutaggregateproductionfunctions,seeAnwarShaikh,"LawsofProduction
and Laws of Algebra:The HumbugProductionFunction,"Reviewof Economicsand
Statistics,vol. 56(February1974),pp. 115-20.

(4) Q = AGa (1 + G')O2 K*P L*y PJ e(x* + Lc)t
with
OX1(X2 , 7,t , > 0, 8 < 0,
where
Q = totaloutputperunitof purchasedproductionlaborinputs
A = positiveconstant
G = level of aggregatecapacityutilization
G' = indexof rateof changeof thatutilization
K* = utilized,nonobsolescentcapitalinputsperhourofpurchased
productionlaborinputs
L* = amountof effectivelaborinputsperhourof purchasedlabor
inputs
= indexof therelativepriceof externalinputs
C = indexof thelevel of innovativepressureonfirmsto improve
productiveefficiency.
Comparativehypothesesaboutthe sourcesof theproductivityslow-
downrequirethatwe movefromthisexpressionfordeterminantsof the
levelofaggregatelaborproductivitytoafocusonchangesinproductivity
overtime.Wedenotex as thelogarithmicrateof changeofX perunitof
time, substitutefromexpressions1and2 into4, andintroduceYiandY2
as coefficientsof adjustmentmediatingthe two determinantsof work
intensity. The following generalexpressionfor changes in aggregate
laborproductivityovertimeis thusobtained:
(5) q X*+ Otlg + ot2Ag + 3k*+ yle[J S D(U, R)]
+ Y2m(W!, B, M*) + 8px + ILC,
with
X, al, X2, , Y1 _Y29F > 09 8 < 0-
We next providea specificationandeconometricestimationof our
generalsocialmodelofproductivitygrowth,as summarizedbyequation
5. We also compareits explanatorypowerwiththatof modelsderived
fromthe moreconventionaltechnicalexplanationsof the productivity
slowdown.
MODEL SPECIFICATION
We estimatethe modelwithannualdatafor the postwaryearsfrom
1948to 1979becausetheseendpointswerebusinesscycle peaksandwe

sought to estimate over the full rangeof completedbusiness cycles
duringthe postwarperiodfor whichall necessarydatawere available.
(Completeannual-but not quarterly-data seriesareavailableforour
desiredvariablesonlybetween1948and1979.)Wefocusonthenonfarm
privatebusiness sector in the United States, excludingboth the farm
andgovernmentsectorsonthegroundsthattheirdynamicsandproduc-
tionprocesses are not adequatelycapturedby the modeldevelopedin
theprevioussection.Wereviewourspecificationofeachofthevariables
inourgeneralsocialmodelintheorderinwhichtheyappearinequation
5. (See the appendixforfulldocumentationof thedatasources.)When
we introducevariables,we normallydefinethemfirstin termsof their
levels, referringbacktotheoriginalexpressionforthelevelofaggregate
productivityinequation4, andthentransformthemintoratesofchange
forestimationof equation5.
Hourly Output. We specify the dependentvariableas the rate of
changeof realoutputperhourof production-workeremployment.To
obtainthis measureof hourlyoutput,we use basic indexes of nonfarm
businessoutputandtotalhoursby the Bureauof LaborStatistics,and
adjusttotalhoursinthedenominatorby theratioof production-worker
employmentto totalemployment."
Wedefinetheproductivitymeasureinrelationto production-worker
hours,ratherthantotalworkerhours,inorderto focus on the differing
activitiesof directlyproductivelabor,ontheonehand,andsurveillance
orsupervisorylabor,on theother.(Weuse standardBLSdatato define
productionworkersas "production"employeesinthegoods-producing
sectorsandas "nonsupervisory"employeesinthe restof the nonfarm
privatebusiness sector. Workersin thiscombinedcategorycomprised
31. Some economistsare skepticalof "value-added"measuresof productivityand
preferto limitthemselvesto "physical"measuresof outputinthenumerator.Thelatter
are availablein the United States only throughthe FederalReserve Boardseries on
manufacturingoutput. Althoughwe recognize that value measures introducesome
potential"noise" intothe measureof totaloutput,we feel comfortablewiththe value-
addedmeasuresfor the purposesof this paperfor two reasons. First, thereare some
obvious advantagesto being able to develop analysesfor the entirenonfarmprivate
businesssector, butphysicalmeasuresof outputareavailableonly for manufacturing.
Second,andmuchmoreimportant,itdoes notappearthatthephysicalandvalue-added
series tell a substantiallydifferentstory for the manufacturingsector for the postwar
economy.Foroneusefulcomparisonofthetwoseriesformanufacturing,seeTomMichl,
"The Lowdownon the Slowdown"(New School for Social Research,Departmentof
Economics,1982).

81.3 percent of total private employmentin 1980.)This choice of a
denominatorfor the outputvariablerecognizesthatthe specificnature
of the contributionof supervisorylaborto productionis qualitatively
differentfromthatof productionworkers;its contributionlies substan-
tiallyin extractingworkfromdirectlyproductiveworkersratherthan
directlytransformingraw materialsand intermediategoods into final
outputs. We show below that our empiricalresults are robust with
respectto thisandalternativespecificationsof thedependentvariable.
CapacityUtilization. TheBureauof EconomicAnalysisdataon the
ratioof actualto potentialGNP, as adjustedperiodicallyby theCouncil
of Economic Advisers, are used to measureeconomy-wide capacity
utilization.Thisratiotrackstheyear-to-yearmovementsofthebusiness
cycleandisusedtomeasuretheimpactofthosefluctuationsonaggregate
laborproductivity.32
CapitalIntensity. The fixed-capitalstock in the nonfarmbusiness
sectoristakenfromtherecentlypublishedBureauofEconomicAnalysis
data.Conformingto ourdefinitionof hourlyoutputandof productive-
laborinputs,wedividethecapitalstockbyestimatedproduction-worker
hours to derive an estimate of the capital-laborratio. Because only
utilizedcapitalinputsaugmenthourlyproductivity,weadjustthecapital
stock by the rateof capacityutilization(as definedabove), withKuas
the utilizedcapital-laborratio. This proceduremeansthat the rate of
changeofcapacityutilizationentersthemodeltwice-once directlyand
once as a componentof themeasureof therateof changeintheutilized
capital-laborratio;italsomeanswe haveto interpretthecoefficientson
both variablescarefully.Thisformulationis nonethelesspreferredfor
the purposesof initialestimationbecause it moredirectlyassesses the
contributionoffluctuationsinthecapitalservicesactuallyappliedinthe
productionprocessthana measureof theunadjustedcapital-laborratio
ora measureof the ratioof theownedcapitalstockto potentialGNP.33
Again, we examine the sensitivity of the results to this particular
formulationin subsequentestimations.
32. Wealsoconsideredandexploredtwoalternativemeasuresof capacityutilization
andcyclicalvariation-the FederalReserveBoard'sseriesoncapacityutilizationandthe
aggregateunemploymentrate.Thedisadvantageof theformeris thatit is onlyavailable
formanufacturing,andnotforthetotalnonfarmprivatebusinesssector;thedisadvantage
ofthelatteris thatitreflectseffectsofbothlaborsupplyanddemand.
33. Wesharesomeoftheskepticismof Cambridge,England,abouttheplausibilityof
aggregatemeasuresofthevalueofheterogeneouscapitalgoods,butwe suspectthatthese

Theissueof therelativeobsolescenceof capitalis morecomplicated,
andthereareno clearguidelinesinthe availableliterature.34Following
the interestingsuggestionsof MartinNeil Baily, we have formulated
severalprovisionalspecificationsofthehypothesisthatthecontribution
of capitalservices dependson variationsin thedegreeof obsolescence
ofthecapitalstock,butwediscusstheseadditionaltestsinthesucceeding
sectionbecauseof theirspeculative(andlargelyinconclusive)nature.
WorkIntensity.WespecifytheMarxeffectthroughdirectestimation
of allcomponentsof L*includedinequation5forwhichit was possible
to obtainannualempiricaldata.
Theeffectivenessof employercontrolis afunctionofthreevariables,
J, S, andD. Relyingon separateworkby JulietB. SchorandSamuel
Bowles, we measureJ, the cost ofjob loss, by a compositemeasureof
theexpectedincomeloss resultingfromjob termination:35
(6) J = Ud [(W - WR)(W + WS)],
where
Ud = averageunemploymentdurationforjob losers, expressedas
afractionof a year
W = averageproduction-workerweeklyafter-taxearnings
criticismsaremorerelevantforthedebatesaboutthemarginalproductivityofcapitalthan
forthe econometricinvestigationof alternativehypothesesconcerningthe productivity
slowdown.As long as the biases in aggregatevalue measuresof the capitalstock are
relativelyconsistentover time, for example,thenchangesin the rateof growthof the
measuredcapitalstock, andthereforein ourmeasureof capitalintensity,are likelyto
provideplausibleindicationsof changesin the physicalratiosof capitalgoods to labor
hoursand, consequently,of changesin the contributionof capitalgoods to the produc-
tivenessof laborhoursovertime.
We also consideredformulatingourvariablefor capitalintensityas the ratioof the
ownedcapitalstocktopotentialoutput,withaneyeontrendsinpotential(butnotactually
utilized)capitalintensity.Thisseemsto usa theoreticallyinferiorspecificationsinceitis
distortedby lagsin the adjustmentof investment(andthereforechangesin the valueof
thecapitalstock)to changesinthetrendgrowthof bothoutputandproductivity;it does
notdirectlymeasurethevalueofthecapitalstock,whichenhanceslaborproductivity,but
focuses insteadon the capitalthatmightaugmentit. Wetestedthis specificationin our
estimationofequation8,inanycase,andthecapitalintensityvariablebecamestatistically
insignificant.
34. SeethediscussioninBaily,"ProductivityandtheServicesofCapitalandLabor";
andBarryP. Bosworth,"CapitalFormationandEconomicPolicy,"BPEA,2:1982,pp.
286-90.
35. See JulietB. SchorandSamuelBowles, "Conflictin the EmploymentRelation
andCostofJobLoss."

WR = averageweeklyincome-replacingbenefitsreceivedby those
withoutwork,includinga weightedallocationof unemploy-
ment compensation, public assistance, food stamps, and
medicalcarebenefits,to reflectrelativeworkereligibilityfor
thosedifferentstreamsof benefits
Ws = averageweekly nonincome-replacingbenefits, particularly
includingeducationandmedicalbenefits.
Definedin thisway, J measuresthe fractionof a year'soverallincome
(includingbothwagesandgovernmentexpenditures)thata workercan
expecttoforgoas aresultofjob loss.
The intensityof supervision,S, is measuredas the ratioof nonpro-
duction-workerhourstoproduction-workerhoursinthenonfarmprivate
businesssector.36
We measure D, the probabilityof dismissal should a worker be
detectedperformingata level of intensitybelow managementexpecta-
tions, by specifyingthe two componentsidentifiedin equation 1. We
define U as the percentageof the nonagriculturallaborforce that is
unionized.WedefineR astheratiooftrendproduction-workeremploy-
menttoitscurrentlevels;thegreateristheratiooftrendtoactuallevels,
themoreslackthereis inthecurrentproduction-workerlaborforceand
thereforethe easier it is to replace a productionworkerwho is not
cooperativeon thejob.
36. Manywhohaveheardus presentourworkhavequestionedtherelevanceof this
measureof theintensityof supervisiononthegroundsthatitincludestoo manytechnical
workerstowarrantitsuseasaproxyforsupervisorylabor.Werelyonthedatadistinction
instandardBureauofLaborStatisticscompilationsbetweenproductionandnonproduction
employeesin the goods sectorsandnonsupervisoryand supervisoryemployeesin the
service sectors. It is true that some of the employees designatedby officialdata as
nonproductionor supervisorypersonnel are not managers,clerical supervisors, or
foremen,butbyfarthelargestproportiondoindeedfallintothosethreecategories.
In 1980,forexample,therewere 13.9millionnonproductionandsupervisoryworkers
on privatenonfarmpayrollsas definedby the BLSdefinitionswe applyin ourstatistical
work.Inthatsameyear,accordingtothemoredetailedthree-digitoccupationalcategories
oftheU.S. BureauoftheCensus,therewereapproximately11.5millionmanagers,clerical
supervisors,andblue-collarsupervisorsintheprivatesector.(Thisapproximationassumes
thatthe sameproportionsof workersworkedin those categoriesin boththe publicand
privatesectors.)Evenin 1980,therefore,at the endof a longperiodof rapidgrowthin
technicalemploymentit remainedtruethatat least 83 percentof workersin this data
categorywereexplicitlyclassifiedas managersandsupervisors.

In the absenceof compellingtheoreticalguidelines,a multiplicative
formforD is arbitrarilyadopted:
(7) D = R /(Lp!Lp,)!U,
whereU is definedas aboveandLpmeasuresthetimetrendof produc-
tion-workerhours,Lp,, for 1948-79.37
Inthispaperwe focus on two of the threevariablescontainedin the
functionalexpressionforchangesinworkers'motivation,m,inequation
5.Wemeasurew!bytherateofchangeinworkers'realspendablehourly
earnings,hypothesizingthatworkers'motivationis higher,as is their
workintensity,the morerapidlytheirtake-homepay increases.38We
express an imperfectproxyforB, the index of the qualityof working
conditions,as the inverseof Z, the industrialaccidentrate;thusB =
Z-1. We postulatean inverse relationbetween the accident rate and
workers'motivation.Accordingtothisformulation,risingaccidentrates
arelikely to erode the qualityof workingconditions,reduceworkers'
motivation,andthereforediminishworkintensity.39
Wehadhopedto findanannualmeasureof workers'job satisfaction
to serve as a proxyfor M*, whichalso affectsworkermotivation.We
couldnotfindasuitableannualseriesfortheentireperiodofobservation
andthuswereunableto includea measureof M*inourequation.40
37. Weplanto exploretheeffectsof a varietyof possiblespecificationsinthefuture;
forthemoment,sincethevariableJcarriesthegreateststatisticalweightintheeconometric
applicationsreportedbelow, we retainthis simplemultiplicativedefinitionandsave for
othereffortsamorecarefulanalysisofthemostappropriateformofinteractionamongthe
twoconstituentvariablesofD. Wetakethesquarerootof theexpressioninequation7to
avoidartificialamplificationofthevarianceofthecompositevariable.
38. Thishypothesisseemsparticularlyrelevantforunionizedproductionworkersin
the UnitedStatesin thepostwarperiodbecauseit correspondsto the logicof collective
bargainingoverthedistributionof thedividendsfromproductivitygrowth-to whathas
usuallybeencalled,indeed,productivitybargaining.
39. One could alternativelyhypothesizethat the accident rate primarilyreflects
speedupon thejob andthata risingaccidentratewouldthereforebe accompaniedby an
increaseinworkintensity,notadecrease.TheactualestimatedresultsfortheeffectsofZ
shouldhelpto discriminatebetweenthese two possibilities:a negativeandstatistically
significanteffectofZonhourlyoutputwouldsuggestthatthenegativeimpactofaspeedup
onworkermotivationswampsanydirectpositiveeffectonworkintensity,whileapositive
effectonproductivitywouldindicatetheopposite.
40. Wenotewithinterest,however,thatthe seriesonjob satisfactionavailableon a
consistentthoughintermittentbasissincethemid-1950sreportstrendsthatareparallelto
trendsin productivitygrowthandsupportourhypothesesaboutspreadingdisaffection
sincethe 1960s.Seethefirstmajorsectionofthispaperandnote7above.

T. Weisskopf,S. Bowles, andD. Gordon 407
External Input Costs. We beginby measuringthe relativepriceof
externalinputcosts by Pf, an index of the relativepriceof fuels. This
choice reflectsthe importanceplacedon oil prices in the mainstream
literature.41Wearenotcontentwiththismeasure,however.
A varietyof externalinputsexists whose supplycan be subjectto
shocksandwhoserelativepricemightthereforeincreaseandadversely
affectthe productivityof otherfactorsof production.Forthe domestic
privatenonfarmbusinesssectorinthepostwarperiod,suchshockshave
occurredin the productionand pricingof many nonagriculturalraw
materials,notjust oil andcoal. This has been due to bothforeignand
domesticeffects. Internationally,OPECandothersuppliershavecom-
binedto producesharpincreasesin the priceof importedoil anda few
otherimportedcrudematerials.Domestically,popularresistanceto the
physicalandenvironmentaleffects of energyproduction-embodied in
the movementfor environmentalregulation,the campaignagainstnu-
clearpower, andthe mineworkers'strugglesfor greatersafety in the
mines-have combined to increase the relative productioncosts of
domesticnonagriculturalcrudematerials.Thesetwodevelopmentshave
jointly affected the relativecosts of externalinputsconsumedby the
domesticprivatenonfarmbusinesseconomy.42
We capture this effect with an alternativevariable representing
externalinputprices,Px, measuredastherelativecostofnonagricultural
crudematerials,whichwe calculatebydividinganindexofthepricesof
fuelsandothercrudematerials(excludingfoodstuffsandfeedstuffs)by
theaggregateGDPpricedeflator.
Theperformanceofthesetwoalternativemeasuresofrelativeexternal
inputpricesis comparedthroughouttherestoftheanalysis.Evenatthis
41. For a summaryof muchof this literaturesee ErnstR. Berndt,"EnergyPrice
Increasesandthe ProductivitySlowdownin UnitedStatesManufacturing,"in Federal
Reserve Bank of Boston, The Decline in Productivity Growth, pp. 60-89. Technically it
oughttobepossibletocapturethesubstitutioneffectofrelativeexternalinputpriceshocks
bymeasuringchangesintherelativequantitiesofenergyinputs,ratherthanintheirprices,
andto rely on pricemeasuresto test for the effect of measurementbiasresultingfrom
doubledeflationof gross outputmeasures(see note 29 above). In practice,however,
quantitymeasuresdonotprovideadditionalexplanatorypowerintheeconometricresults
reportedbelow, so thatwe areconstrainedforempiricalreasonsto relyexclusivelyon
relativepricemeasuresof thiseffect.See thetextbelowforadiscussionof theeffectof a
quantitymeasureof energyinputs.
42. See thediscussionin Bowles, Gordon,andWeisskopf,Beyondthe WasteLand,
pp.91-94and136-38.

stageof initialspecification,the principaldifferencesareevident.The
index of relative fuel prices pays attentionto only one category of
externalinputsandthereforetendsto confuse a particularinstanceof
recentpriceshockwiththemoregeneralpoliticaleconomicproblemsof
sluggish supply responses, occupational safety, and environmental
problems.Thefuelscoveredintheindexof relativefuelprices,indeed,
accountforonlyaboutone-third(byvalueweights)of thetotalvalueof
inputsincludedin the nonagriculturalcrude-materialpriceindex. This
index, because of its focus on the priceof importedoil, also tends to
givetoomuchattentiontoasingletypeofforeignshock-the successive
OPECpricehikes-and too littletootherdomesticmovementsthatalso
affectrelativeinputpricesof crudematerials.4
Innovative Pressure on Business. Following Schumpeter's clues
aboutcreativedestruction,we focus on the rateof businessfailuresas
anindirectmeasureof thepressureonbusinessfirmsto remaincompet-
itive andadoptavailabletechnicalinnovations.Datafor the business-
failurerateareavailableon a continuousbasisthroughoutthe postwar
period.Use of thisvariabledoes notimplythatbusinessfailuresarethe
onlyormostimportantcauseofproductiveinnovationbutthatvariations
in the business-failureratereflectandthemselvesdirectlycovarywith
awidevarietyof factorsaffectingbusinessinnovationforwhichcontin-
uousannualdataarenotavailablethroughoutthepostwarperiod.
Furtherapplyingthesuppositionthatvariationsininnovativepressure
havecumulativeeffects, we transformthebusiness-failurerateto elimi-
nateshort-termcyclicalvariationandtohighlightitsmoreseculartrends.
Weresidualizethebusinessfailurerateonthelevelofcapacityutilization
and then take a three-year (end-of-period)moving average of this
residualizedvariable.We labelthis variableC*. This involves the hy-
pothesis thatrecentvariationsin the business-failureratethatare not
simplyfunctionsof fluctuationsin the level of capacityutilizationcan
providea proxymeasureof recenttrendsinthe intensityof innovative
pressureon firms.44Wefurthertest the robustnessof this specification
43. We placeheavyemphasisin ourbookon the roleof domesticandinternational
"popularresistance"as a sourceof risingexternalinputpricesafterthemid-1960s.This
involvesanargumentaboutandaninterpretationof ourvariableforexternalinputprices
thatcannotbefurtherexploredthroughtheeconometricsof thispaper;see ibid., chaps.
4and6.
44. Inbackgroundworkforthispaperwe developeda modelof the failureratethat
supportsthis interpretation.We showthat,whenone controlsforthe level of capacity

by exploringseveralalternativevariablesbasedon thebusiness-failure
rate.
EMPIRICAL ESTIMATION
Basedon these variabledefinitions,we arriveat a finalspecification
of equation5 forthepurposesof initialeconometricestimation:
(8) q =A* +Otlg + t2Ag +f3k,, + Pyle
+ 'y2IAw! + 'Y22Z+ bPx + pC*,
where the variablesare defined as above. We hypothesize that the
coefficientsX*, xtl,(X2, ,Yi lY21, andp.arepositiveandthatY22 and 8 are
negative.Asinstandardeconometricestimation,weassumeastochastic
componentinthedependentvariable,withu, distributedwithzeromean
andC2 variance.45
Two considerationsinfluencedthe empiricalwork: first, since we
could not assumea priorithatourhypothesesaboutthe effects of our
unfamiliarandunconventional"social" variableswereassociatedwith
changes in aggregateproductivity,it seemed crucialto test these hy-
pothesesagainstthenullhypothesisof no significantstatisticalassocia-
tion.46Furthermore,although we preferredto estimate equation 8
directly,ratherthanuse a modifiedversionof theaggregateproduction
functionwithoutputas the dependentvariableanda measureof labor
utilizationand the cost of capital,variationsin the failurerate are almostcompletely
explainedby trendsincapitalintensity,anda trendtermthatclosely mirrorsthepattern
reflectedin figure2-suggesting a smoothandinvertedU-shapetrajectoryin the deter-
minantsof thefailurerateinthepostwarperiod.
Wechosea three-yearperiodforthemovingaverageof thefailureratebecausethisis
thelengthof theperiodoverwhichpastvaluesof theratearesignificantinexplainingits
currentvaluesinanautoregressiveestimation.
45. Weuseordinaryleastsquaresestimationinthetestsreportedhere,asimplechoice
thatdoes notappearto requireemendationas a resultof the specificationor theresults
themselves.Seethetextbelowforcommentsonpotentialsimultaneousequationbiasand
autocorrelation.
We add here the moreinclusivemeasureof externalinputprices,p,, because we
believe it bettercapturesthe politicaleconomicdeterminantsof variationsin external
inputprices.Belowwe compareitsperformancewiththenarrowermeasure,pf.
46. Oneof the mostglaringpresumptionsof the accounting-methodapproachis its
axiomaticapplicationofthemarginal-productivitytheoryofwagestotheadjustmentsfor
"quality"in estimatinglabor.See thecriticaldiscussionof thisprocedureinthe section
entitled"ExtensionsandCompetingHypotheses"below.

410 Brookings Papers oniEconomic Activity, 2:1983
hourson the right-handside of the equation,we test the sensitivityof
ourresultsto thisparticularspecification.47
We show below the averagevaluesfor the dependentvariable,the
averageannualgrowthin realoutputperproduction-workerhour,for
the entireperiod, 1948-79, andfor three subperiodsin which, as the
previousliteratureattests, the productivityslowdownis most clearly
revealed.48These dataareconsistentwithotherestimatesin the litera-
ture. Differences in estimates of the timing and magnitudeof the
productivityslowdownprimarilyarein the levels of subperiodgrowth
rates.If we compareourestimateswiththose of MartinNeil Baily,for
example, the productivitygrowthrates estimatedhere are higherfor
everysubperiod,buttherelativemagnitudesofdeclinefromoneperiod
to the next areroughlycomparable;the highergrowthratesestimated
herearedueto the use of production-workerhoursas the denominator
inthemeasureofhourlyoutput,sincetotalworkerhoursgrewatamore
rapidratethanproduction-workerhoursduringthepostwarperiod.49
Average annual
rate of growth
(percent)
1948-79 2.34
1948-66 2.87
1966-73 2.19
1973-79 0.85
47. Ourpreferredspecificationfocusesdirectlyontheobjectofanalysis-variationin
the rateof changeof productivity-and poses a morestringenttest forthe explanatory
powerof thevariousindependentvariablesintheanalysis.
Onedangerof this specification,of course,is thatwe maycompoundtheproblemof
potentialerrorsinvariablesbyincludinglaborhoursinthedenominatoron boththeleft-
andright-handsidesof the equation.Wetest forthispossiblebiasbelowby comparing
ourbasicresultswitha specificationinwhichonlyoutputappearsontheleft-handsideof
theequationandlaborhoursappearsasanindependentvariable.
48. Productivitygrowthis definedastheaverageannuallogarithmicrateofgrowthof
outputper production-workerhour.The averagesare calculatedas the means of the
individualannualgrowthrates,sincethesearetheactualobservationsforthedependent
variablein subsequenteconometricestimation.See Bowles, Gordon,and Weisskopf,
Beyondthe WasteLand,chap.2, forfurtherjustificationoftheseperiodsforcomparison.
49. See Baily, "Productivityandthe Servicesof CapitalandLabor,"p. 9. Growth
ratesforthecounterparttoourdependentvariabledefinedasoutputpertotal-workerhour
fortheentireperiodandthethreesubperiodsreportedintable1were2.08,2.56,2.10,and
0.63percent,respectively.

Table 1. A Social Model of Aggregate Productivity Growth in U.S. Nonfarm Private
Business, 1948-79a
Equation
Independentvariable 1-1 1-2 1-3 1-4 1-5 1-6 1-7
Trend growth rate, X* 0.015 0.016 0.016 0.004 0.015 0.017 0.017
(7.98) (6.35) (7.97) (1.02) (6.92) (8.07) (8.39)
Capacity utilization
g 0.485 0.519 0.449 0.419 0.408 0.442 0.415
(4.80) (4.43) (4.68) (4.58) (3.47) (4.48) (4.41)
Ag 0.074 0.049 0.082 0.094 0.091 0.125 0.103
(1.33) (0.78) (1.48) (1.74) (1.50) (2.29) (1.96)
Utilized capital-labor ratio, 0.401 0.385 0.411 0.419 0.472 0.373 0.373
k,, (4.64) (4.26) (4.74) (4.99) (5.29) (4.17) (4.26)
Effectiveness of employers' 0.035 ... ... ... ... ... ...
control, e (3.74)
Real spendable earnings, 0.075 0.073 0.072 0.069 0.121 0.292b
Avw! (1.68) (1.57) (1.60) (1.57) (2.71) (1.37)
Accident rate, z -0.083 -0.088 -0.086 -0.087 -0.100 -0.081
(-3.50) (-3.38) (-3.60) (-3.74) (-3.29) (-3.23)
Relative extemal input -0.040 -0.042 -0.041 -0.037 - 0.013b -0.048 -0.055
prices, p, (-2.13) (-2.13) (-2.14) (-1.95) (-0.52) (-2.62) (-3.27)
Business failure measure, 0.011 0.011 0.011 0.020b 0.012 0.010 0.008
C* (2.81) (2.72) (2.75) (2.98) (2.43) (2.50) (2.20)
Cost of job loss,j . . . 0.036 0.034 0.031 0.027 0.027 0.034
(3.72) (3.60) (3.41) (2.74) (2.48) (3.58)
Supervision intensity, s ... 0.001 ... ... ... ... .
(0.01)
Probability of dismissal, d ... 0.100 ... ... .. . ... ...
(1.20)
Predicted job satisfaction, ... ... ... ... ... ... 0.450
i! !(3.74)
Summary statistic
1R2 0.913 0.908 0.910 0.914 0.893 0.906 0.909
Standard error of
estimate 0.005 0.005 0.005 0.005 0.005 0.005 0.005
Durbin-Watson 2.095 1.891 2.092 2.200 2.184 2.214 2.096
Source: Authors' estimates based on data cited in the appendix.
a. The dependent variable is the rate of change in output per production-worker hours, measured as change in
logs. The independent variables are all growth rates measured as change in logs, with the exception of C*, which is
a level variable, and Ag and Aw!, which are first differences of changes in logs. Given first differences, the actual
observations are for 1949-79. The numbers in parentheses are t-statistics.
b. The values for C* (equation 1-4), p, (equation 1-5), and Au'! (equation 1-6) are for the alternative variables
described in the text.
Wepresentintable 1theresultsof thedirectestimationof equation8
andseveralvariantsofit. Webegininequation1-1withthebasicmodel.
All coefficientshave the expected signs. All independentvariablesare
statistically significant(on one-tailed tests), with six variables at 1
percent, two at 5 percent, and one, Ag, at 10 percent. The variance
explainedby the model is quite high given its focus on changes in

412 Br-ookingsPaper-s on Econiomic Activity, 2:1983
aggregateproductivity(ratherthan levels), and the Durbin-Watson
statisticallowsusto acceptthenullhypothesisof zeroautocorrelation.
The second columnpresentsresultsfor the samemodelwithe, the
effectivenessofemployercontrol,decomposedintoitsthreeconstituent
variables.The model remainsrobust,with virtuallyno changein the
coefficientson theothersignificantindependentvariables.Thecost-of-
job-loss componentof e, the variablej,is highlysignificantandclearly
constitutestheunderlyingcomponentof e withthegreatestexplanatory
power;its standardizedregressioncoefficientis actuallyhigherthanthe
standardizedregressioncoefficientfore inequation1-1.Thecoefficient
ondalsohastherightsign,butotherwiseperformspoorly.Thecoefficient
for s, the rateof changeof the intensityof supervision,is statistically
insignificant.
Wepresentinthe columnfor 1-3the resultsof equation8 estimated
withonlyj includedas a measureof variationsin e. As inequation1-2,
j is positive and significant,with a coefficientnearto the value for e
reportedin 1-1,andtheresultsfortheothervariablesremainconsistent.
Giventheessentialequivalenceof the resultsin 1-1,1-2,and1-3,we
choose to work with the simplerversion presentedin equation 1-3.
Althoughwe believe that the version in 1-1, with the fully specified
variablee, representsthe theoreticallycorrect representationof the
microeconomiclogic of our analysisof the effectiveness of employer
control,we retainthe versionin 1-3for subsequentstatisticalanalysis
to avoid the possibility that some other components of e with less
powerful statistical effects might complicate these comparativeand
forecastingexercises.50
We turn next to our proxy measure for innovative pressure on
50. See Bowles, "CompetitiveWageDetermination,"forfurthertheoreticaljustifi-
cationfortheuseofJ, ourmeasureof thecostofjobloss.
Somereadersof anearlydrafthaveexpressedconcernaboutapossibleidentityinthe
relationbetweenQandJ inourmodelbecausetheexpressionforJ includesthewagerate
andthereis a close relationbetweentheratesof changeof productivityandthe product
wage(see equation6). Thisconcernis misplaced,however,becauseWappearsin both
thenumeratorandthedenominatorofJ. IftheotherdeterminantsofJ wereconstantand
small,J wouldfluctuatearoundavalueof 1.00,ratherthanaroundthevalueof W.Infact,
variationsinJ aredominatedby variationsin ourmeasuresof unemploymentduration,
income-replacingpayments,andnonincome-replacingpayments,notbychangesineither
theproductwageorinourmeasureofrealspendablehourlyearnings.Indeed,thesimple
correlationcoefficientbetweentherateof changeof the productwageandJ is actually
negative.

business. Assumingfor the momentthatthe business-failureratecon-
stitutes the appropriatebasis for such a measure, we test several
alternativespecificationsof this variable.We reportin 1-4the results
witha simplealternativespecification,C', anonresidualized,three-year
moving average of the business-failurerate. The magnitudeof the
coefficientsfor C* and C' differbecause the residualizationshifts the
unitsofmeasurementonC*,butthestandardizedregressioncoefficients
arealmostexactlyequalandso aretheirlevelsof statisticalsignificance.
This suggests, in other words, that our results are insensitive to the
choice aboutwhetheror not to adjustthe business-failureratefor its
cyclicalcovariationwiththecapacityutilizationrate;we prefersuchan
adjustmentto focus attentionon longer-termtrendsin innovativepres-
sure.Wealsotestafive-yearmovingaverageofthebusiness-failurerate
anda specificationin whichwe treatthe currentvalueof this rateas a
two-yearautoregressivedistributedlagonpastvaluesofthesamerate-
analternativemeasureof the notionof cumulativetrendsininnovative
pressure.Bothof these specifications,notreportedhere,yieldedequiv-
alentresultson boththe innovativepressurevariableandon the other
independentvariablesinthemodel.5"
Weconsidernextourspecificationof thevariablemeasuringrelative
externalinputprices.Equation1-5revealstheresultsof substitutingpf,
thevariablemeasuringrelativefuelprices,forp. Althoughitscoefficient
hastherightsign,itis quitesmallandstatisticallyinsignificant.If, aswe
and many other economists suspect, movements in one or another
componentof externalinputpriceshavedampenedproductivitygrowth
since the mid-1960s,our measure of relative nonagriculturalcrude-
materialspricesappearsto capturethiseffectmoreadequatelythanthe
narrowermeasureof relativefuelprices.
51. We also exploreda definitionof the variablebasedon net businessformations,
takinginto accountnot only failuresbutnew enterprises.The problemwithadditional
attentionto new enterpriseformation,however,is thatit is almostentirelya functionof
realinterestratesandappearsto havelittleconnectionwithotherdimensionsof macro-
economicactivity. The frequencyof net businessformationsgrew steadilyduringthe
postwarperiod,particularlyduringthe yearsof low realinterestratesin the 1970s,and
does notappearto havemovedinclose relationto otherfactorslikelyto affectaggregate
productivity.Failuresandnetbusinessformations,indeed,moveinoppositedirections;
the simplecorrelationbetweenthe frequencyof businessfailuresandnet formationsis
- 0.47.Wethereforethinkthatthefailureratemorecloselyreflectstheunderlyingfactors
thataffectrelativesuccessatmodernizationandinnovation.

414 Br-ookinigsPaper-s oniEconomic Activ'ity, 2:1983
The two other "social" variables,zAw!and z, also deserve further
comment.Althoughproductivitygrowthandwage growthare closely
linked,theintroductionof siw!, one of ourindicatorsof workermotiva-
tion, shouldnotintroducesimultaneousequationbiasintoourequation
becauseitrelatesthelevelof workermotivation(andthereforethelevel
of work intensity and of aggregatelaborproductivity)to the rate of
changeinworkers'take-homepayinequation8. Thereis no necessary
statisticalrelationbetweenthefirstandsecondderivativesofavariable,
so thereis no apriorireasonto treatsiv'! asanendogenousvariableand
toexpectsimultaneousequationbiasasaresultof itsinclusionin8. The
simplecorrelationcoefficientbetween q and svit! is only 0.37, barely
morethanhalfthesimplecorrelationbetweenqanduv!,thefirstderivative
of realspendablehourlyearnings.
Theresultsreportedforequation1-3maystillreflectcausalityrunning
fromtherateofgrowthofproductivitytotherateofgrowthofrealwages
(andtherefore,if even indirectly,ZAw!).Wecantest for this possibility
by residualizingAi'! on the rateof growthof the productwage (hourly
compensationdeflatedby the GDP deflator)and then includingthat
residualized variable, &w!* in 8. (This residualization allows us to
eliminate the portion of variations in sit'! that are plausibly understood
asendogenouslydeterminedalongwithq.)Wepresenttheresultsofthis
procedure in equation 1-6, in which we substitutethe residualized
independentvariable zw!* for Zw!. There is some decrease in the
coefficientandstatisticalsignificanceofthismeasureofourmotivational
hypothesis, but the variableremainssignificantat 10percentand the
rest of the results remainrobust. (Thereis little changein the results
largelybecausethe simplecorrelationbetweenzAw!andrateof change
of theproductwageis only0.36.)52
We would obviously have preferreda measure of the quality of
workingconditionsthatcovers theentirenonfarmprivatesectorrather
thanjust the industrialaccidentrate, z. Lackingsuch a series, we are
reasonablyconfidentthat trends in manufacturingclosely paralleled
trendsinothersectors.Duringtheperiodwhenindustrialaccidentrates
52. We further estimated Phillips curve relations with uw!and Aeit!, respectively, as
dependentvariables,andthelaggedunemploymentrateandproductivitygrowth,among
others,as independentvariables.Whilesucha modelexplainsa largeportionof annual
variationin wt,!-therateofgrowthofourmeasureof realspendableearnings-it explains
only 10percentof theannualvariationinAit!(adjustedfordegreesoffreedom).

T. Weisskopf, S. Bowles, anidD. Gor-don 415
beganto riseafterthemid-1960s,forexample,therewasbothrisingjob
dissatisfactionandincreasingconcernaboutworkingconditionsamong
theworkersinwhite-collaroccupations.53
Thereis one finaltest of the adequacyof Aivt!andz as measuresof
factorsaffectingworkermotivation.As noted at the beginningof this
paper,one consistentdataseries on employee satisfactionis available
since the mid-1950s;thereare six observationson this series over the
periodofouranalysis(seenote7).Despitethelimiteddegreesoffreedom,
ourtwo proxyvariablesshouldcertainlyhelpexplainvariationsin the
intermittentvariablemeasuringjob satisfaction.Whenwe regressM!,
this measureof job satisfaction,on W!andZ, indeed, the two proxy
variableshavetheexpectedsignsandexplain65percentofthevariation
inM!. Wecanfurtherusethecoefficientsfromthisregressiontogenerate
anannualseriesof predictedjob satisfactionbasedontheannualvalues
for w! andZ. Equation1-7shows the resultsof substitutingthe rateof
changeof this measureof predictedjob satisfactionfor zAw!and z in
equation1-3.Thecoefficienton in! is statisticallysignificantandallthe
otherresultsremainrobust.Thisfurtherstrengthenstheconclusionthat
theeffects of &i'!! andz in equations1-1through1-3captureto a large
degreethe effects of shiftsin factorsaffectingworkermotivation.We
retainthose variablesfor subsequentanalysis,ratherthanourmeasure
of predictedjob satisfaction,because the coefficientsgeneratingthat
measurearebasedon only six observations.
The three columnsof table 2 providetests of the sensitivityof the
results to the specificationof the dependentvariable. Equation2-1
replicates1-3,exceptthattotal employeehours,ratherthanproduction-
workerhours, are in the denominatorof productivity,the dependent
variable(and, maintainingconsistency, in the indexof k14,the ratioof
utilizedcapitalto labor).Themodelremainsrobust,withno significant
changesincoefficientmagnitudeor significancebetween 1-3and2-1.
The results of decomposing the dependent variable specified in
equation8appearinequation2-2.Thedependentvariablein2-2isy, the
changeinoutputinthenonfarmprivatebusinesssector;andthechange
inproduction-workeremployment,4p,is addedto theright-handsideas
an additionalindependentvariable.We also convert the measureof
53. See Qualityof WorkLife(InstituteforSocialResearch,Universityof Michigan,
1979);andthesummariesinStaines, 'IsWorkerDissatisfactionRising?"

Table 2. Alternative Dependent Variables in the Social Model of Productivity Growth
in U.S. Nonfarm Private Business, 1948-79a
Independent Equation
variable 2-1 2-2 2-3
0.014 0.026 0.016
(7.78) (4.97) (8.09)
g 0.481 0.799
(4.84) (4.09)
Ag 0.091 0.056 0.120
(1.61) (1.05) (4.31)
ku, 0.417b 0.143b 0.427
(4.69) (0.92) (5.11)
j 0.033 0.024 0.028
(3.52) (2.38) (4.77)
AW! 0.080 0.089 0.064
(1.74) (2.06) (1.47)
z -0.079 -0.068 -0.083
(-3.25) (-2.81) (-3.55)
p." -0.037 -0.022 -0.040
(-1.91) (-1.10) (-2.12)
C* 0.011 0.012 0.010
(2.64) (3.17) (2.67)
Ip ... 0.619
(3.28)
Summary statistic
R2 0.917 0.983 0.899
Standard error of estimate 0.005 0.004 0.005
Durbin-Watson 2.254 2.402 2.142
Source: Authors' estimates based on data cited in the appendix.
a. The dependent variable in 2-1 is the rate of change of output per total-employee hour; in 2-2, the rate of change
of output; and in 2-3, the rate of change of production-worker hourly output residualized on the rate of change of
capacity utilization. The variable Ipis rate of change in production-worker hours. Given first differences, actual years
of observation are 1949-79. The numbers in parentheses are t-statistics.
b. Adjusted to correspond with the redefinition of the dependent variable, as described in the text.
capitalintensityto a directmeasureof therateof changeof theutilized
capitalstock by removinglaborhoursfromthe denominator.TheR2is
naturally higher in 2-2 than in 1-3.54 Otherwise the model is relatively
insensitiveto this changein the formof specificationof the dependent
variable:the social variables are robust, althoughj falls slightly in
magnitudeof coefficientandstatisticalsignificance;thetechnicalvaria-
bles are the ones, interestingly,that shift substantiallywith this re-
54. Theresultsin2-1areroughlyconsistentwiththeassumptionof constantreturns
to scaleimpliedintheinitialspecificationof equations4 and5, forthecoefficienton Ipis
close to theshareof productionworkers'compensationinnonfarmbusinessincome.

specification,inasmuchas the coefficientson g, ku,andp, all change
greatly.
Because manyof the variablesincludedin the modelvaryover the
businesscycle, it seems importantto test the sensitivityof ourresults
thusfarby usinga dependentvariablethatis itselfcyclicallyadjusted.
Wecomputedsucha variableby firstregressingq, therateof changeof
productivity,on g, the rateof changeof capacityutilization,andthen
takingthe residualfromthat equationas our measureof decyclicized
productivitygrowth. We then regressedthis measureof residualized
productivitygrowth on all variablesin equation8 with the obvious
exceptionofg itself.Wereporttheseresultsinequation2-3.55Giventhe
possibilities of cyclical covariationamong many of the independent
variables,theresultsaresurprisinglyrobust.TheR2remainshigh.Only
zAg,thechangeintherateofchangeofcapacityutilization,isdramatically
affected by moving to the decyclicized dependentvariable, with its
coefficientincreasinginbothmagnitudeandsignificance.
COMPARATIVE PERFORMANCE
We retainthe decyclicized specificationin 2-3 for the purposesof
comparativetests of the model's explanatorypower and structural
stability.Table3comparesourresultswiththosethatmightbeextracted
fromprevailingdiscussionsof theproductivityslowdown.Equation3-1
reproducesthe model as reportedin 2-3. Equation3-2 estimates the
simplestpossibletechnicalmodelthatcanbederivedfromtheavailable
mainstreamliterature;itincludeszAgto controlfor sluggishfirmadjust-
mentsto changesinthe level of businessactivity;k11,the changeinthe
utilized capital-laborratio, reflectingthe concern of the mainstream
55. Therearea varietyof waysof performingthisresidualization;one canworkfrom
eitherequation4 or5, whichhavecomparableresults.Twofeaturesof thedecyclicized
modelin 2-3 areworthnotingin this regard.First,the fact thatthe effects of] remain
robustevenafterthedependentvariablehasbeenresidualizedong suggeststhatitseffects
arenot simplyanadditionalvariablereflectingthecycle. Second,we havetestedforthe
relativeimportanceofthetwomaincomponentsofJ inequation6:unemploymentduration
andrelativeincomeloss. Whenwe reestimate2-3withthe ratesof changeof those two
componentssubstitutedfor j, both componentvariablesare statisticallysignificant,
althoughthe standardizedregressioncoefficienton theunemployment-durationcompo-
nent is the largerof the two. All otherresultsremainessentiallyunchangedwith this
decomposition.

418 Br0ookingsPapers on Economic Activity, 2:1983
Table 3. Alternative Models of Aggregate Productivity Growth in U.S. Nonfarm Private
Business, 1948-79a
Independent Equation
variable 3-1 3-2 3-3 3-4
0.016 0.009 0.011 0.015
(8.09) (2.12) (2.50) (7.09)
Ag 0.120 0.048 0.066 0.114
(4.31) (0.91) (1.25) (3.74)
kl, 0.427 0.758 0.771 0.469
(5.11) (4.57) (4.50) (5.39)
j 0.028 ... ... 0.028
(4.77) (4.72)
Aw!! 0.064 . . . . .. 0.075
(1.47) (1.67)
z -0.083 ... ... -0.076
(-3.55) (-2.73)
P, -0.040 ... ... -0.039
(-2.12) (-2.04)
C* 0.010 . . . . . . 0.011
(2.67) (2.49)
Pf . . -0.037 0.028 . . .
(-1.06) (0.48)
Dumtny,967-73 . . . . . -0.009 -0.002
(-1.82) (-1.04)
Duxmmy1974-79 . . . ... -0.011 0.001
(-1.21) (0.25)
Summary statistic
R2 0.899 0.485 0.512 0.899
Standard error of estimate 0.005 0.010 0.010 0.005
Durbin-Watson 2.142 1.517 1.767 2.257
Source: Authors' estimates based on sources cited in the appendix.
a. The dependent variable is the logarithmic rate of change in output per production-worker hour residualized on
the logarithmic rate of change in capacity utilization. The variable pf denotes the rate of change in relative fuel
prices. Given first differences, the actual years of observation are 1949-79. The numbers in parentheses are
t-statistics.
literaturewith capitalintensityandthe hypothesisof capitalshortage;
andpr' the measureof relative inputprices that seems to reflectthe
hypothesesofrecentdiscussionsabouttheoil-priceshockmostclosely.
We considerelaborationsandextensionsof this simpleformulationof
thetechnicalmodelinthefollowingsection,withparticularattentionto
theeffectsof changesincharacteristicsof thelaborforce.
Several conclusions flow from the comparisonof the results in
equations3-1and3-2.First,thiskindof"technical"modelthatexcludes
oursocialvariablesneitherexplainsas muchof the annualvariationin

(decyclicized) productivitygrowth nor remains as free of potential
inefficiencyfromautocorrelationas the moreinclusivemodelreported
in 3-1. Second, such a technicalmodelgeneratessubstantiallyhigher
estimates of the impact of variationsin utilized capital intensity on
productivity;assuming,on both theoreticalandeconometricgrounds,
thatthe socialmodelis amorefullyspecifiedequation,thissuggeststhat
conventionalstudies have sufferedfrom underspecificationbias and
havesubstantiallyoverestimatedtheimpactof slowdownsintherateof
growthof the capital-laborratio. Third,p, is statisticallyinsignificant
even in the morelimitedtechnicalmodelandappearsagainas a more
statisticallyimprecisemeasureof effectsof externalinputprices.
Thereis anotherusefulway to assess econometricallythe abilityof
these alternativemodels to accountfor period-specificdeclines in the
rateof productivitygrowth.As thedatainthedisplayabovereport,the
rate of growthof outputper production-workerhour slowed by 0.68
percentagepoints from 1948-66 to 1966-73 and by 2.02 points from
1948-66to 1973-79.Thecorrespondingdecrementsforthedecyclicized
measureof productivitygrowthwe use, the dependentvariablein 3-1
and3-2,were0.69and1.76percentagepoints,respectively.Wecantest
for the ability of these two models to capturethese between-period
declines by addingdummyvariablesfor the years between 1967and
1973and between 1974and 1979.If the other independentvariables
accountnot only for annualvariationsin productivitygrowthbut also
foraveragebetween-perioddeclines,theestimatedcoefficientsonthese
respective dummy variables should be statisticallyinsignificantand
close to zero. If, on the otherhand,one oranotherof the modelscould
account for little of a specific between-perioddecline, the estimated
coefficient on that dummyvariablewould be close to the values of
- 0.0069or - 0.0176and,presumably,statisticallysignificant.
Equations3-3and3-4showresultsof thistest forthetwo models.In
3-3,the resultsforthe technicalmodel,the 1967-73dummyvariableis
statisticallysignificantat 5 percent and is actually greaterthan the
between-perioddifferenceinproductivitygrowthrates;thisconfirmsa
familiarresult in conventionalstudies of the slowdownfrom 1966to
1973-that is, capitalintensitygrewveryrapidlyduringthisperiod,and
technicalfactors are unableto account for any of the first phase of
slower productivitygrowth between 1966 and 1973. The estimated
coefficienton the second dummyvariablein 3-3 fallsjust below a 10

420 Brookings Papers oniEconomic Activity, 2:1983
percentlevel of significanceandis equalto roughlythree-fifthsof the
overalldeclineinproductivitygrowthfrom1948-66to 1974-79,further
confirmingthe recent conclusionsof Bosworthand many others that
slower capitalformationin the 1970sexplains relativelylittle of the
productivityslowdowninthoseyears.56
Equation3-4 suggests, in contrast,thatthe basic social modelpre-
sented in 3-1 can account statisticallyfor virtuallyall between-period
productivityslowdowns; neither dummy variablereported in 3-4 is
statisticallysignificant,andbotharequitesmall.
What factors in that model play the greatest (estimated)role in
accountingfor the slowdown itself? We can providedirect statistical
estimatesof therelativeexplanatorypowerof thedifferentconstitutent
variablesrepresentedin equation8. Relyingon the generalalgebraic
result, in matrix notation, zA^= PAX, we can estimate the predicted
changein q betweenthe variousphasesof the postwarperiodthatwe
expect to result from the movements in the respective independent
variablesusingcoefficientscalculatedinthebasicestimationofequation
8.Wepresentthiscomparisonintable4;therewe accountforthedecline
in q fromthe 1948-66boom years to the 1966-73phase of slowdown
andthenagainforthedeclineingrowthratesfrom1948-66to 1973-79.
Wemakeuse of thecompleteversionof themodelreportedinequation
1-3,ratherthanthedecyclicizedversionin2-3,inorderto takeaccount
ofdifferencesintherateofchangeofcapacityutilizationfromtheperiod
of prosperitytothe successive periodsof slowdown.
The firstandthirdcolumnspresentthe predictedpercentagepoints
of decline in hourlyproductivitygrowth,while the second andfourth
columns report the percentageof the total predicteddecline that is
attributableto each of the variables.For explainingthe decline from
1948-66to 1973-79we use thecoefficientsreportedincolumn1-3.For
thedeclinefrom1948-66to 1966-73we use coefficientsforthecompa-
rable equationestimatedfrom the shorterperiodfrom 1948to 1973.
Furtherto clarify the relativecontributionsof differentvariableswe
providesubtotalsby the majorcategoriesof factors stipulatedin the
originalpresentationof the socialmodelof productivitygrowth.57
56. See Bosworth,"CapitalFormationandEconomicPolicy."
57. Theprocedureappliedintable4 wouldnotbe validif a significantnumberof the
variablesincorporatedin the calculationswere statisticallyinsignificant;since we are
relyingonlyonvariablesthatproveto be statisticallysignificantincolumn1-3,ouruseof

Weisskopf1983

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Andere mochten auch

Andere mochten auch (20)

Ähnlich wie Weisskopf1983

Ähnlich wie Weisskopf1983 (20)

Mehr von economicgrowthcucea

Mehr von economicgrowthcucea (20)

Weisskopf1983